1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/super.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  *
7  *  super.c contains code to handle: - mount structures
8  *                                   - super-block tables
9  *                                   - filesystem drivers list
10  *                                   - mount system call
11  *                                   - umount system call
12  *                                   - ustat system call
13  *
14  * GK 2/5/95  -  Changed to support mounting the root fs via NFS
15  *
16  *  Added kerneld support: Jacques Gelinas and Bjorn Ekwall
17  *  Added change_root: Werner Almesberger & Hans Lermen, Feb '96
18  *  Added options to /proc/mounts:
19  *    Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
20  *  Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
21  *  Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
22  */
23 
24 #include <linux/export.h>
25 #include <linux/slab.h>
26 #include <linux/blkdev.h>
27 #include <linux/mount.h>
28 #include <linux/security.h>
29 #include <linux/writeback.h>		/* for the emergency remount stuff */
30 #include <linux/idr.h>
31 #include <linux/mutex.h>
32 #include <linux/backing-dev.h>
33 #include <linux/rculist_bl.h>
34 #include <linux/cleancache.h>
35 #include <linux/fsnotify.h>
36 #include <linux/lockdep.h>
37 #include <linux/user_namespace.h>
38 #include "internal.h"
39 
40 static int thaw_super_locked(struct super_block *sb);
41 
42 static LIST_HEAD(super_blocks);
43 static DEFINE_SPINLOCK(sb_lock);
44 
45 static char *sb_writers_name[SB_FREEZE_LEVELS] = {
46 	"sb_writers",
47 	"sb_pagefaults",
48 	"sb_internal",
49 };
50 
51 /*
52  * One thing we have to be careful of with a per-sb shrinker is that we don't
53  * drop the last active reference to the superblock from within the shrinker.
54  * If that happens we could trigger unregistering the shrinker from within the
55  * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
56  * take a passive reference to the superblock to avoid this from occurring.
57  */
super_cache_scan(struct shrinker * shrink,struct shrink_control * sc)58 static unsigned long super_cache_scan(struct shrinker *shrink,
59 				      struct shrink_control *sc)
60 {
61 	struct super_block *sb;
62 	long	fs_objects = 0;
63 	long	total_objects;
64 	long	freed = 0;
65 	long	dentries;
66 	long	inodes;
67 
68 	sb = container_of(shrink, struct super_block, s_shrink);
69 
70 	/*
71 	 * Deadlock avoidance.  We may hold various FS locks, and we don't want
72 	 * to recurse into the FS that called us in clear_inode() and friends..
73 	 */
74 	if (!(sc->gfp_mask & __GFP_FS))
75 		return SHRINK_STOP;
76 
77 	if (!trylock_super(sb))
78 		return SHRINK_STOP;
79 
80 	if (sb->s_op->nr_cached_objects)
81 		fs_objects = sb->s_op->nr_cached_objects(sb, sc);
82 
83 	inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
84 	dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
85 	total_objects = dentries + inodes + fs_objects + 1;
86 	if (!total_objects)
87 		total_objects = 1;
88 
89 	/* proportion the scan between the caches */
90 	dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
91 	inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
92 	fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
93 
94 	/*
95 	 * prune the dcache first as the icache is pinned by it, then
96 	 * prune the icache, followed by the filesystem specific caches
97 	 *
98 	 * Ensure that we always scan at least one object - memcg kmem
99 	 * accounting uses this to fully empty the caches.
100 	 */
101 	sc->nr_to_scan = dentries + 1;
102 	freed = prune_dcache_sb(sb, sc);
103 	sc->nr_to_scan = inodes + 1;
104 	freed += prune_icache_sb(sb, sc);
105 
106 	if (fs_objects) {
107 		sc->nr_to_scan = fs_objects + 1;
108 		freed += sb->s_op->free_cached_objects(sb, sc);
109 	}
110 
111 	up_read(&sb->s_umount);
112 	return freed;
113 }
114 
super_cache_count(struct shrinker * shrink,struct shrink_control * sc)115 static unsigned long super_cache_count(struct shrinker *shrink,
116 				       struct shrink_control *sc)
117 {
118 	struct super_block *sb;
119 	long	total_objects = 0;
120 
121 	sb = container_of(shrink, struct super_block, s_shrink);
122 
123 	/*
124 	 * We don't call trylock_super() here as it is a scalability bottleneck,
125 	 * so we're exposed to partial setup state. The shrinker rwsem does not
126 	 * protect filesystem operations backing list_lru_shrink_count() or
127 	 * s_op->nr_cached_objects(). Counts can change between
128 	 * super_cache_count and super_cache_scan, so we really don't need locks
129 	 * here.
130 	 *
131 	 * However, if we are currently mounting the superblock, the underlying
132 	 * filesystem might be in a state of partial construction and hence it
133 	 * is dangerous to access it.  trylock_super() uses a SB_BORN check to
134 	 * avoid this situation, so do the same here. The memory barrier is
135 	 * matched with the one in mount_fs() as we don't hold locks here.
136 	 */
137 	if (!(sb->s_flags & SB_BORN))
138 		return 0;
139 	smp_rmb();
140 
141 	if (sb->s_op && sb->s_op->nr_cached_objects)
142 		total_objects = sb->s_op->nr_cached_objects(sb, sc);
143 
144 	total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
145 	total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
146 
147 	if (!total_objects)
148 		return SHRINK_EMPTY;
149 
150 	total_objects = vfs_pressure_ratio(total_objects);
151 	return total_objects;
152 }
153 
destroy_super_work(struct work_struct * work)154 static void destroy_super_work(struct work_struct *work)
155 {
156 	struct super_block *s = container_of(work, struct super_block,
157 							destroy_work);
158 	int i;
159 
160 	for (i = 0; i < SB_FREEZE_LEVELS; i++)
161 		percpu_free_rwsem(&s->s_writers.rw_sem[i]);
162 	kfree(s);
163 }
164 
destroy_super_rcu(struct rcu_head * head)165 static void destroy_super_rcu(struct rcu_head *head)
166 {
167 	struct super_block *s = container_of(head, struct super_block, rcu);
168 	INIT_WORK(&s->destroy_work, destroy_super_work);
169 	schedule_work(&s->destroy_work);
170 }
171 
172 /* Free a superblock that has never been seen by anyone */
destroy_unused_super(struct super_block * s)173 static void destroy_unused_super(struct super_block *s)
174 {
175 	if (!s)
176 		return;
177 	up_write(&s->s_umount);
178 	list_lru_destroy(&s->s_dentry_lru);
179 	list_lru_destroy(&s->s_inode_lru);
180 	security_sb_free(s);
181 	put_user_ns(s->s_user_ns);
182 	kfree(s->s_subtype);
183 	free_prealloced_shrinker(&s->s_shrink);
184 	/* no delays needed */
185 	destroy_super_work(&s->destroy_work);
186 }
187 
188 /**
189  *	alloc_super	-	create new superblock
190  *	@type:	filesystem type superblock should belong to
191  *	@flags: the mount flags
192  *	@user_ns: User namespace for the super_block
193  *
194  *	Allocates and initializes a new &struct super_block.  alloc_super()
195  *	returns a pointer new superblock or %NULL if allocation had failed.
196  */
alloc_super(struct file_system_type * type,int flags,struct user_namespace * user_ns)197 static struct super_block *alloc_super(struct file_system_type *type, int flags,
198 				       struct user_namespace *user_ns)
199 {
200 	struct super_block *s = kzalloc(sizeof(struct super_block),  GFP_USER);
201 	static const struct super_operations default_op;
202 	int i;
203 
204 	if (!s)
205 		return NULL;
206 
207 	INIT_LIST_HEAD(&s->s_mounts);
208 	s->s_user_ns = get_user_ns(user_ns);
209 	init_rwsem(&s->s_umount);
210 	lockdep_set_class(&s->s_umount, &type->s_umount_key);
211 	/*
212 	 * sget() can have s_umount recursion.
213 	 *
214 	 * When it cannot find a suitable sb, it allocates a new
215 	 * one (this one), and tries again to find a suitable old
216 	 * one.
217 	 *
218 	 * In case that succeeds, it will acquire the s_umount
219 	 * lock of the old one. Since these are clearly distrinct
220 	 * locks, and this object isn't exposed yet, there's no
221 	 * risk of deadlocks.
222 	 *
223 	 * Annotate this by putting this lock in a different
224 	 * subclass.
225 	 */
226 	down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
227 
228 	if (security_sb_alloc(s))
229 		goto fail;
230 
231 	for (i = 0; i < SB_FREEZE_LEVELS; i++) {
232 		if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
233 					sb_writers_name[i],
234 					&type->s_writers_key[i]))
235 			goto fail;
236 	}
237 	init_waitqueue_head(&s->s_writers.wait_unfrozen);
238 	s->s_bdi = &noop_backing_dev_info;
239 	s->s_flags = flags;
240 	if (s->s_user_ns != &init_user_ns)
241 		s->s_iflags |= SB_I_NODEV;
242 	INIT_HLIST_NODE(&s->s_instances);
243 	INIT_HLIST_BL_HEAD(&s->s_roots);
244 	mutex_init(&s->s_sync_lock);
245 	INIT_LIST_HEAD(&s->s_inodes);
246 	spin_lock_init(&s->s_inode_list_lock);
247 	INIT_LIST_HEAD(&s->s_inodes_wb);
248 	spin_lock_init(&s->s_inode_wblist_lock);
249 
250 	s->s_count = 1;
251 	atomic_set(&s->s_active, 1);
252 	mutex_init(&s->s_vfs_rename_mutex);
253 	lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
254 	init_rwsem(&s->s_dquot.dqio_sem);
255 	s->s_maxbytes = MAX_NON_LFS;
256 	s->s_op = &default_op;
257 	s->s_time_gran = 1000000000;
258 	s->cleancache_poolid = CLEANCACHE_NO_POOL;
259 
260 	s->s_shrink.seeks = DEFAULT_SEEKS;
261 	s->s_shrink.scan_objects = super_cache_scan;
262 	s->s_shrink.count_objects = super_cache_count;
263 	s->s_shrink.batch = 1024;
264 	s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
265 	if (prealloc_shrinker(&s->s_shrink))
266 		goto fail;
267 	if (list_lru_init_memcg(&s->s_dentry_lru, &s->s_shrink))
268 		goto fail;
269 	if (list_lru_init_memcg(&s->s_inode_lru, &s->s_shrink))
270 		goto fail;
271 	return s;
272 
273 fail:
274 	destroy_unused_super(s);
275 	return NULL;
276 }
277 
278 /* Superblock refcounting  */
279 
280 /*
281  * Drop a superblock's refcount.  The caller must hold sb_lock.
282  */
__put_super(struct super_block * s)283 static void __put_super(struct super_block *s)
284 {
285 	if (!--s->s_count) {
286 		list_del_init(&s->s_list);
287 		WARN_ON(s->s_dentry_lru.node);
288 		WARN_ON(s->s_inode_lru.node);
289 		WARN_ON(!list_empty(&s->s_mounts));
290 		security_sb_free(s);
291 		put_user_ns(s->s_user_ns);
292 		kfree(s->s_subtype);
293 		call_rcu(&s->rcu, destroy_super_rcu);
294 	}
295 }
296 
297 /**
298  *	put_super	-	drop a temporary reference to superblock
299  *	@sb: superblock in question
300  *
301  *	Drops a temporary reference, frees superblock if there's no
302  *	references left.
303  */
put_super(struct super_block * sb)304 static void put_super(struct super_block *sb)
305 {
306 	spin_lock(&sb_lock);
307 	__put_super(sb);
308 	spin_unlock(&sb_lock);
309 }
310 
311 
312 /**
313  *	deactivate_locked_super	-	drop an active reference to superblock
314  *	@s: superblock to deactivate
315  *
316  *	Drops an active reference to superblock, converting it into a temporary
317  *	one if there is no other active references left.  In that case we
318  *	tell fs driver to shut it down and drop the temporary reference we
319  *	had just acquired.
320  *
321  *	Caller holds exclusive lock on superblock; that lock is released.
322  */
deactivate_locked_super(struct super_block * s)323 void deactivate_locked_super(struct super_block *s)
324 {
325 	struct file_system_type *fs = s->s_type;
326 	if (atomic_dec_and_test(&s->s_active)) {
327 		cleancache_invalidate_fs(s);
328 		unregister_shrinker(&s->s_shrink);
329 		fs->kill_sb(s);
330 
331 		/*
332 		 * Since list_lru_destroy() may sleep, we cannot call it from
333 		 * put_super(), where we hold the sb_lock. Therefore we destroy
334 		 * the lru lists right now.
335 		 */
336 		list_lru_destroy(&s->s_dentry_lru);
337 		list_lru_destroy(&s->s_inode_lru);
338 
339 		put_filesystem(fs);
340 		put_super(s);
341 	} else {
342 		up_write(&s->s_umount);
343 	}
344 }
345 
346 EXPORT_SYMBOL(deactivate_locked_super);
347 
348 /**
349  *	deactivate_super	-	drop an active reference to superblock
350  *	@s: superblock to deactivate
351  *
352  *	Variant of deactivate_locked_super(), except that superblock is *not*
353  *	locked by caller.  If we are going to drop the final active reference,
354  *	lock will be acquired prior to that.
355  */
deactivate_super(struct super_block * s)356 void deactivate_super(struct super_block *s)
357 {
358         if (!atomic_add_unless(&s->s_active, -1, 1)) {
359 		down_write(&s->s_umount);
360 		deactivate_locked_super(s);
361 	}
362 }
363 
364 EXPORT_SYMBOL(deactivate_super);
365 
366 /**
367  *	grab_super - acquire an active reference
368  *	@s: reference we are trying to make active
369  *
370  *	Tries to acquire an active reference.  grab_super() is used when we
371  * 	had just found a superblock in super_blocks or fs_type->fs_supers
372  *	and want to turn it into a full-blown active reference.  grab_super()
373  *	is called with sb_lock held and drops it.  Returns 1 in case of
374  *	success, 0 if we had failed (superblock contents was already dead or
375  *	dying when grab_super() had been called).  Note that this is only
376  *	called for superblocks not in rundown mode (== ones still on ->fs_supers
377  *	of their type), so increment of ->s_count is OK here.
378  */
grab_super(struct super_block * s)379 static int grab_super(struct super_block *s) __releases(sb_lock)
380 {
381 	s->s_count++;
382 	spin_unlock(&sb_lock);
383 	down_write(&s->s_umount);
384 	if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
385 		put_super(s);
386 		return 1;
387 	}
388 	up_write(&s->s_umount);
389 	put_super(s);
390 	return 0;
391 }
392 
393 /*
394  *	trylock_super - try to grab ->s_umount shared
395  *	@sb: reference we are trying to grab
396  *
397  *	Try to prevent fs shutdown.  This is used in places where we
398  *	cannot take an active reference but we need to ensure that the
399  *	filesystem is not shut down while we are working on it. It returns
400  *	false if we cannot acquire s_umount or if we lose the race and
401  *	filesystem already got into shutdown, and returns true with the s_umount
402  *	lock held in read mode in case of success. On successful return,
403  *	the caller must drop the s_umount lock when done.
404  *
405  *	Note that unlike get_super() et.al. this one does *not* bump ->s_count.
406  *	The reason why it's safe is that we are OK with doing trylock instead
407  *	of down_read().  There's a couple of places that are OK with that, but
408  *	it's very much not a general-purpose interface.
409  */
trylock_super(struct super_block * sb)410 bool trylock_super(struct super_block *sb)
411 {
412 	if (down_read_trylock(&sb->s_umount)) {
413 		if (!hlist_unhashed(&sb->s_instances) &&
414 		    sb->s_root && (sb->s_flags & SB_BORN))
415 			return true;
416 		up_read(&sb->s_umount);
417 	}
418 
419 	return false;
420 }
421 
422 /**
423  *	generic_shutdown_super	-	common helper for ->kill_sb()
424  *	@sb: superblock to kill
425  *
426  *	generic_shutdown_super() does all fs-independent work on superblock
427  *	shutdown.  Typical ->kill_sb() should pick all fs-specific objects
428  *	that need destruction out of superblock, call generic_shutdown_super()
429  *	and release aforementioned objects.  Note: dentries and inodes _are_
430  *	taken care of and do not need specific handling.
431  *
432  *	Upon calling this function, the filesystem may no longer alter or
433  *	rearrange the set of dentries belonging to this super_block, nor may it
434  *	change the attachments of dentries to inodes.
435  */
generic_shutdown_super(struct super_block * sb)436 void generic_shutdown_super(struct super_block *sb)
437 {
438 	const struct super_operations *sop = sb->s_op;
439 
440 	if (sb->s_root) {
441 		shrink_dcache_for_umount(sb);
442 		sync_filesystem(sb);
443 		sb->s_flags &= ~SB_ACTIVE;
444 
445 		fsnotify_unmount_inodes(sb);
446 		cgroup_writeback_umount();
447 
448 		evict_inodes(sb);
449 
450 		if (sb->s_dio_done_wq) {
451 			destroy_workqueue(sb->s_dio_done_wq);
452 			sb->s_dio_done_wq = NULL;
453 		}
454 
455 		if (sop->put_super)
456 			sop->put_super(sb);
457 
458 		if (!list_empty(&sb->s_inodes)) {
459 			printk("VFS: Busy inodes after unmount of %s. "
460 			   "Self-destruct in 5 seconds.  Have a nice day...\n",
461 			   sb->s_id);
462 		}
463 	}
464 	spin_lock(&sb_lock);
465 	/* should be initialized for __put_super_and_need_restart() */
466 	hlist_del_init(&sb->s_instances);
467 	spin_unlock(&sb_lock);
468 	up_write(&sb->s_umount);
469 	if (sb->s_bdi != &noop_backing_dev_info) {
470 		bdi_put(sb->s_bdi);
471 		sb->s_bdi = &noop_backing_dev_info;
472 	}
473 }
474 
475 EXPORT_SYMBOL(generic_shutdown_super);
476 
477 /**
478  *	sget_userns -	find or create a superblock
479  *	@type:	filesystem type superblock should belong to
480  *	@test:	comparison callback
481  *	@set:	setup callback
482  *	@flags:	mount flags
483  *	@user_ns: User namespace for the super_block
484  *	@data:	argument to each of them
485  */
sget_userns(struct file_system_type * type,int (* test)(struct super_block *,void *),int (* set)(struct super_block *,void *),int flags,struct user_namespace * user_ns,void * data)486 struct super_block *sget_userns(struct file_system_type *type,
487 			int (*test)(struct super_block *,void *),
488 			int (*set)(struct super_block *,void *),
489 			int flags, struct user_namespace *user_ns,
490 			void *data)
491 {
492 	struct super_block *s = NULL;
493 	struct super_block *old;
494 	int err;
495 
496 	if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) &&
497 	    !(type->fs_flags & FS_USERNS_MOUNT) &&
498 	    !capable(CAP_SYS_ADMIN))
499 		return ERR_PTR(-EPERM);
500 retry:
501 	spin_lock(&sb_lock);
502 	if (test) {
503 		hlist_for_each_entry(old, &type->fs_supers, s_instances) {
504 			if (!test(old, data))
505 				continue;
506 			if (user_ns != old->s_user_ns) {
507 				spin_unlock(&sb_lock);
508 				destroy_unused_super(s);
509 				return ERR_PTR(-EBUSY);
510 			}
511 			if (!grab_super(old))
512 				goto retry;
513 			destroy_unused_super(s);
514 			return old;
515 		}
516 	}
517 	if (!s) {
518 		spin_unlock(&sb_lock);
519 		s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
520 		if (!s)
521 			return ERR_PTR(-ENOMEM);
522 		goto retry;
523 	}
524 
525 	err = set(s, data);
526 	if (err) {
527 		spin_unlock(&sb_lock);
528 		destroy_unused_super(s);
529 		return ERR_PTR(err);
530 	}
531 	s->s_type = type;
532 	strlcpy(s->s_id, type->name, sizeof(s->s_id));
533 	list_add_tail(&s->s_list, &super_blocks);
534 	hlist_add_head(&s->s_instances, &type->fs_supers);
535 	spin_unlock(&sb_lock);
536 	get_filesystem(type);
537 	register_shrinker_prepared(&s->s_shrink);
538 	return s;
539 }
540 
541 EXPORT_SYMBOL(sget_userns);
542 
543 /**
544  *	sget	-	find or create a superblock
545  *	@type:	  filesystem type superblock should belong to
546  *	@test:	  comparison callback
547  *	@set:	  setup callback
548  *	@flags:	  mount flags
549  *	@data:	  argument to each of them
550  */
sget(struct file_system_type * type,int (* test)(struct super_block *,void *),int (* set)(struct super_block *,void *),int flags,void * data)551 struct super_block *sget(struct file_system_type *type,
552 			int (*test)(struct super_block *,void *),
553 			int (*set)(struct super_block *,void *),
554 			int flags,
555 			void *data)
556 {
557 	struct user_namespace *user_ns = current_user_ns();
558 
559 	/* We don't yet pass the user namespace of the parent
560 	 * mount through to here so always use &init_user_ns
561 	 * until that changes.
562 	 */
563 	if (flags & SB_SUBMOUNT)
564 		user_ns = &init_user_ns;
565 
566 	/* Ensure the requestor has permissions over the target filesystem */
567 	if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) && !ns_capable(user_ns, CAP_SYS_ADMIN))
568 		return ERR_PTR(-EPERM);
569 
570 	return sget_userns(type, test, set, flags, user_ns, data);
571 }
572 
573 EXPORT_SYMBOL(sget);
574 
drop_super(struct super_block * sb)575 void drop_super(struct super_block *sb)
576 {
577 	up_read(&sb->s_umount);
578 	put_super(sb);
579 }
580 
581 EXPORT_SYMBOL(drop_super);
582 
drop_super_exclusive(struct super_block * sb)583 void drop_super_exclusive(struct super_block *sb)
584 {
585 	up_write(&sb->s_umount);
586 	put_super(sb);
587 }
588 EXPORT_SYMBOL(drop_super_exclusive);
589 
__iterate_supers(void (* f)(struct super_block *))590 static void __iterate_supers(void (*f)(struct super_block *))
591 {
592 	struct super_block *sb, *p = NULL;
593 
594 	spin_lock(&sb_lock);
595 	list_for_each_entry(sb, &super_blocks, s_list) {
596 		if (hlist_unhashed(&sb->s_instances))
597 			continue;
598 		sb->s_count++;
599 		spin_unlock(&sb_lock);
600 
601 		f(sb);
602 
603 		spin_lock(&sb_lock);
604 		if (p)
605 			__put_super(p);
606 		p = sb;
607 	}
608 	if (p)
609 		__put_super(p);
610 	spin_unlock(&sb_lock);
611 }
612 /**
613  *	iterate_supers - call function for all active superblocks
614  *	@f: function to call
615  *	@arg: argument to pass to it
616  *
617  *	Scans the superblock list and calls given function, passing it
618  *	locked superblock and given argument.
619  */
iterate_supers(void (* f)(struct super_block *,void *),void * arg)620 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
621 {
622 	struct super_block *sb, *p = NULL;
623 
624 	spin_lock(&sb_lock);
625 	list_for_each_entry(sb, &super_blocks, s_list) {
626 		if (hlist_unhashed(&sb->s_instances))
627 			continue;
628 		sb->s_count++;
629 		spin_unlock(&sb_lock);
630 
631 		down_read(&sb->s_umount);
632 		if (sb->s_root && (sb->s_flags & SB_BORN))
633 			f(sb, arg);
634 		up_read(&sb->s_umount);
635 
636 		spin_lock(&sb_lock);
637 		if (p)
638 			__put_super(p);
639 		p = sb;
640 	}
641 	if (p)
642 		__put_super(p);
643 	spin_unlock(&sb_lock);
644 }
645 
646 /**
647  *	iterate_supers_type - call function for superblocks of given type
648  *	@type: fs type
649  *	@f: function to call
650  *	@arg: argument to pass to it
651  *
652  *	Scans the superblock list and calls given function, passing it
653  *	locked superblock and given argument.
654  */
iterate_supers_type(struct file_system_type * type,void (* f)(struct super_block *,void *),void * arg)655 void iterate_supers_type(struct file_system_type *type,
656 	void (*f)(struct super_block *, void *), void *arg)
657 {
658 	struct super_block *sb, *p = NULL;
659 
660 	spin_lock(&sb_lock);
661 	hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
662 		sb->s_count++;
663 		spin_unlock(&sb_lock);
664 
665 		down_read(&sb->s_umount);
666 		if (sb->s_root && (sb->s_flags & SB_BORN))
667 			f(sb, arg);
668 		up_read(&sb->s_umount);
669 
670 		spin_lock(&sb_lock);
671 		if (p)
672 			__put_super(p);
673 		p = sb;
674 	}
675 	if (p)
676 		__put_super(p);
677 	spin_unlock(&sb_lock);
678 }
679 
680 EXPORT_SYMBOL(iterate_supers_type);
681 
__get_super(struct block_device * bdev,bool excl)682 static struct super_block *__get_super(struct block_device *bdev, bool excl)
683 {
684 	struct super_block *sb;
685 
686 	if (!bdev)
687 		return NULL;
688 
689 	spin_lock(&sb_lock);
690 rescan:
691 	list_for_each_entry(sb, &super_blocks, s_list) {
692 		if (hlist_unhashed(&sb->s_instances))
693 			continue;
694 		if (sb->s_bdev == bdev) {
695 			sb->s_count++;
696 			spin_unlock(&sb_lock);
697 			if (!excl)
698 				down_read(&sb->s_umount);
699 			else
700 				down_write(&sb->s_umount);
701 			/* still alive? */
702 			if (sb->s_root && (sb->s_flags & SB_BORN))
703 				return sb;
704 			if (!excl)
705 				up_read(&sb->s_umount);
706 			else
707 				up_write(&sb->s_umount);
708 			/* nope, got unmounted */
709 			spin_lock(&sb_lock);
710 			__put_super(sb);
711 			goto rescan;
712 		}
713 	}
714 	spin_unlock(&sb_lock);
715 	return NULL;
716 }
717 
718 /**
719  *	get_super - get the superblock of a device
720  *	@bdev: device to get the superblock for
721  *
722  *	Scans the superblock list and finds the superblock of the file system
723  *	mounted on the device given. %NULL is returned if no match is found.
724  */
get_super(struct block_device * bdev)725 struct super_block *get_super(struct block_device *bdev)
726 {
727 	return __get_super(bdev, false);
728 }
729 EXPORT_SYMBOL(get_super);
730 
__get_super_thawed(struct block_device * bdev,bool excl)731 static struct super_block *__get_super_thawed(struct block_device *bdev,
732 					      bool excl)
733 {
734 	while (1) {
735 		struct super_block *s = __get_super(bdev, excl);
736 		if (!s || s->s_writers.frozen == SB_UNFROZEN)
737 			return s;
738 		if (!excl)
739 			up_read(&s->s_umount);
740 		else
741 			up_write(&s->s_umount);
742 		wait_event(s->s_writers.wait_unfrozen,
743 			   s->s_writers.frozen == SB_UNFROZEN);
744 		put_super(s);
745 	}
746 }
747 
748 /**
749  *	get_super_thawed - get thawed superblock of a device
750  *	@bdev: device to get the superblock for
751  *
752  *	Scans the superblock list and finds the superblock of the file system
753  *	mounted on the device. The superblock is returned once it is thawed
754  *	(or immediately if it was not frozen). %NULL is returned if no match
755  *	is found.
756  */
get_super_thawed(struct block_device * bdev)757 struct super_block *get_super_thawed(struct block_device *bdev)
758 {
759 	return __get_super_thawed(bdev, false);
760 }
761 EXPORT_SYMBOL(get_super_thawed);
762 
763 /**
764  *	get_super_exclusive_thawed - get thawed superblock of a device
765  *	@bdev: device to get the superblock for
766  *
767  *	Scans the superblock list and finds the superblock of the file system
768  *	mounted on the device. The superblock is returned once it is thawed
769  *	(or immediately if it was not frozen) and s_umount semaphore is held
770  *	in exclusive mode. %NULL is returned if no match is found.
771  */
get_super_exclusive_thawed(struct block_device * bdev)772 struct super_block *get_super_exclusive_thawed(struct block_device *bdev)
773 {
774 	return __get_super_thawed(bdev, true);
775 }
776 EXPORT_SYMBOL(get_super_exclusive_thawed);
777 
778 /**
779  * get_active_super - get an active reference to the superblock of a device
780  * @bdev: device to get the superblock for
781  *
782  * Scans the superblock list and finds the superblock of the file system
783  * mounted on the device given.  Returns the superblock with an active
784  * reference or %NULL if none was found.
785  */
get_active_super(struct block_device * bdev)786 struct super_block *get_active_super(struct block_device *bdev)
787 {
788 	struct super_block *sb;
789 
790 	if (!bdev)
791 		return NULL;
792 
793 restart:
794 	spin_lock(&sb_lock);
795 	list_for_each_entry(sb, &super_blocks, s_list) {
796 		if (hlist_unhashed(&sb->s_instances))
797 			continue;
798 		if (sb->s_bdev == bdev) {
799 			if (!grab_super(sb))
800 				goto restart;
801 			up_write(&sb->s_umount);
802 			return sb;
803 		}
804 	}
805 	spin_unlock(&sb_lock);
806 	return NULL;
807 }
808 
user_get_super(dev_t dev)809 struct super_block *user_get_super(dev_t dev)
810 {
811 	struct super_block *sb;
812 
813 	spin_lock(&sb_lock);
814 rescan:
815 	list_for_each_entry(sb, &super_blocks, s_list) {
816 		if (hlist_unhashed(&sb->s_instances))
817 			continue;
818 		if (sb->s_dev ==  dev) {
819 			sb->s_count++;
820 			spin_unlock(&sb_lock);
821 			down_read(&sb->s_umount);
822 			/* still alive? */
823 			if (sb->s_root && (sb->s_flags & SB_BORN))
824 				return sb;
825 			up_read(&sb->s_umount);
826 			/* nope, got unmounted */
827 			spin_lock(&sb_lock);
828 			__put_super(sb);
829 			goto rescan;
830 		}
831 	}
832 	spin_unlock(&sb_lock);
833 	return NULL;
834 }
835 
836 /**
837  *	do_remount_sb - asks filesystem to change mount options.
838  *	@sb:	superblock in question
839  *	@sb_flags: revised superblock flags
840  *	@data:	the rest of options
841  *      @force: whether or not to force the change
842  *
843  *	Alters the mount options of a mounted file system.
844  */
do_remount_sb(struct super_block * sb,int sb_flags,void * data,int force)845 int do_remount_sb(struct super_block *sb, int sb_flags, void *data, int force)
846 {
847 	int retval;
848 	int remount_ro;
849 
850 	if (sb->s_writers.frozen != SB_UNFROZEN)
851 		return -EBUSY;
852 
853 #ifdef CONFIG_BLOCK
854 	if (!(sb_flags & SB_RDONLY) && bdev_read_only(sb->s_bdev))
855 		return -EACCES;
856 #endif
857 
858 	remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb);
859 
860 	if (remount_ro) {
861 		if (!hlist_empty(&sb->s_pins)) {
862 			up_write(&sb->s_umount);
863 			group_pin_kill(&sb->s_pins);
864 			down_write(&sb->s_umount);
865 			if (!sb->s_root)
866 				return 0;
867 			if (sb->s_writers.frozen != SB_UNFROZEN)
868 				return -EBUSY;
869 			remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb);
870 		}
871 	}
872 	shrink_dcache_sb(sb);
873 
874 	/* If we are remounting RDONLY and current sb is read/write,
875 	   make sure there are no rw files opened */
876 	if (remount_ro) {
877 		if (force) {
878 			sb->s_readonly_remount = 1;
879 			smp_wmb();
880 		} else {
881 			retval = sb_prepare_remount_readonly(sb);
882 			if (retval)
883 				return retval;
884 		}
885 	}
886 
887 	if (sb->s_op->remount_fs) {
888 		retval = sb->s_op->remount_fs(sb, &sb_flags, data);
889 		if (retval) {
890 			if (!force)
891 				goto cancel_readonly;
892 			/* If forced remount, go ahead despite any errors */
893 			WARN(1, "forced remount of a %s fs returned %i\n",
894 			     sb->s_type->name, retval);
895 		}
896 	}
897 	sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (sb_flags & MS_RMT_MASK);
898 	/* Needs to be ordered wrt mnt_is_readonly() */
899 	smp_wmb();
900 	sb->s_readonly_remount = 0;
901 
902 	/*
903 	 * Some filesystems modify their metadata via some other path than the
904 	 * bdev buffer cache (eg. use a private mapping, or directories in
905 	 * pagecache, etc). Also file data modifications go via their own
906 	 * mappings. So If we try to mount readonly then copy the filesystem
907 	 * from bdev, we could get stale data, so invalidate it to give a best
908 	 * effort at coherency.
909 	 */
910 	if (remount_ro && sb->s_bdev)
911 		invalidate_bdev(sb->s_bdev);
912 	return 0;
913 
914 cancel_readonly:
915 	sb->s_readonly_remount = 0;
916 	return retval;
917 }
918 
do_emergency_remount_callback(struct super_block * sb)919 static void do_emergency_remount_callback(struct super_block *sb)
920 {
921 	down_write(&sb->s_umount);
922 	if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
923 	    !sb_rdonly(sb)) {
924 		/*
925 		 * What lock protects sb->s_flags??
926 		 */
927 		do_remount_sb(sb, SB_RDONLY, NULL, 1);
928 	}
929 	up_write(&sb->s_umount);
930 }
931 
do_emergency_remount(struct work_struct * work)932 static void do_emergency_remount(struct work_struct *work)
933 {
934 	__iterate_supers(do_emergency_remount_callback);
935 	kfree(work);
936 	printk("Emergency Remount complete\n");
937 }
938 
emergency_remount(void)939 void emergency_remount(void)
940 {
941 	struct work_struct *work;
942 
943 	work = kmalloc(sizeof(*work), GFP_ATOMIC);
944 	if (work) {
945 		INIT_WORK(work, do_emergency_remount);
946 		schedule_work(work);
947 	}
948 }
949 
do_thaw_all_callback(struct super_block * sb)950 static void do_thaw_all_callback(struct super_block *sb)
951 {
952 	down_write(&sb->s_umount);
953 	if (sb->s_root && sb->s_flags & SB_BORN) {
954 		emergency_thaw_bdev(sb);
955 		thaw_super_locked(sb);
956 	} else {
957 		up_write(&sb->s_umount);
958 	}
959 }
960 
do_thaw_all(struct work_struct * work)961 static void do_thaw_all(struct work_struct *work)
962 {
963 	__iterate_supers(do_thaw_all_callback);
964 	kfree(work);
965 	printk(KERN_WARNING "Emergency Thaw complete\n");
966 }
967 
968 /**
969  * emergency_thaw_all -- forcibly thaw every frozen filesystem
970  *
971  * Used for emergency unfreeze of all filesystems via SysRq
972  */
emergency_thaw_all(void)973 void emergency_thaw_all(void)
974 {
975 	struct work_struct *work;
976 
977 	work = kmalloc(sizeof(*work), GFP_ATOMIC);
978 	if (work) {
979 		INIT_WORK(work, do_thaw_all);
980 		schedule_work(work);
981 	}
982 }
983 
984 static DEFINE_IDA(unnamed_dev_ida);
985 
986 /**
987  * get_anon_bdev - Allocate a block device for filesystems which don't have one.
988  * @p: Pointer to a dev_t.
989  *
990  * Filesystems which don't use real block devices can call this function
991  * to allocate a virtual block device.
992  *
993  * Context: Any context.  Frequently called while holding sb_lock.
994  * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
995  * or -ENOMEM if memory allocation failed.
996  */
get_anon_bdev(dev_t * p)997 int get_anon_bdev(dev_t *p)
998 {
999 	int dev;
1000 
1001 	/*
1002 	 * Many userspace utilities consider an FSID of 0 invalid.
1003 	 * Always return at least 1 from get_anon_bdev.
1004 	 */
1005 	dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1,
1006 			GFP_ATOMIC);
1007 	if (dev == -ENOSPC)
1008 		dev = -EMFILE;
1009 	if (dev < 0)
1010 		return dev;
1011 
1012 	*p = MKDEV(0, dev);
1013 	return 0;
1014 }
1015 EXPORT_SYMBOL(get_anon_bdev);
1016 
free_anon_bdev(dev_t dev)1017 void free_anon_bdev(dev_t dev)
1018 {
1019 	ida_free(&unnamed_dev_ida, MINOR(dev));
1020 }
1021 EXPORT_SYMBOL(free_anon_bdev);
1022 
set_anon_super(struct super_block * s,void * data)1023 int set_anon_super(struct super_block *s, void *data)
1024 {
1025 	return get_anon_bdev(&s->s_dev);
1026 }
1027 EXPORT_SYMBOL(set_anon_super);
1028 
kill_anon_super(struct super_block * sb)1029 void kill_anon_super(struct super_block *sb)
1030 {
1031 	dev_t dev = sb->s_dev;
1032 	generic_shutdown_super(sb);
1033 	free_anon_bdev(dev);
1034 }
1035 EXPORT_SYMBOL(kill_anon_super);
1036 
kill_litter_super(struct super_block * sb)1037 void kill_litter_super(struct super_block *sb)
1038 {
1039 	if (sb->s_root)
1040 		d_genocide(sb->s_root);
1041 	kill_anon_super(sb);
1042 }
1043 EXPORT_SYMBOL(kill_litter_super);
1044 
ns_test_super(struct super_block * sb,void * data)1045 static int ns_test_super(struct super_block *sb, void *data)
1046 {
1047 	return sb->s_fs_info == data;
1048 }
1049 
ns_set_super(struct super_block * sb,void * data)1050 static int ns_set_super(struct super_block *sb, void *data)
1051 {
1052 	sb->s_fs_info = data;
1053 	return set_anon_super(sb, NULL);
1054 }
1055 
mount_ns(struct file_system_type * fs_type,int flags,void * data,void * ns,struct user_namespace * user_ns,int (* fill_super)(struct super_block *,void *,int))1056 struct dentry *mount_ns(struct file_system_type *fs_type,
1057 	int flags, void *data, void *ns, struct user_namespace *user_ns,
1058 	int (*fill_super)(struct super_block *, void *, int))
1059 {
1060 	struct super_block *sb;
1061 
1062 	/* Don't allow mounting unless the caller has CAP_SYS_ADMIN
1063 	 * over the namespace.
1064 	 */
1065 	if (!(flags & SB_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
1066 		return ERR_PTR(-EPERM);
1067 
1068 	sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags,
1069 			 user_ns, ns);
1070 	if (IS_ERR(sb))
1071 		return ERR_CAST(sb);
1072 
1073 	if (!sb->s_root) {
1074 		int err;
1075 		err = fill_super(sb, data, flags & SB_SILENT ? 1 : 0);
1076 		if (err) {
1077 			deactivate_locked_super(sb);
1078 			return ERR_PTR(err);
1079 		}
1080 
1081 		sb->s_flags |= SB_ACTIVE;
1082 	}
1083 
1084 	return dget(sb->s_root);
1085 }
1086 
1087 EXPORT_SYMBOL(mount_ns);
1088 
1089 #ifdef CONFIG_BLOCK
set_bdev_super(struct super_block * s,void * data)1090 static int set_bdev_super(struct super_block *s, void *data)
1091 {
1092 	s->s_bdev = data;
1093 	s->s_dev = s->s_bdev->bd_dev;
1094 	s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
1095 
1096 	return 0;
1097 }
1098 
test_bdev_super(struct super_block * s,void * data)1099 static int test_bdev_super(struct super_block *s, void *data)
1100 {
1101 	return (void *)s->s_bdev == data;
1102 }
1103 
mount_bdev(struct file_system_type * fs_type,int flags,const char * dev_name,void * data,int (* fill_super)(struct super_block *,void *,int))1104 struct dentry *mount_bdev(struct file_system_type *fs_type,
1105 	int flags, const char *dev_name, void *data,
1106 	int (*fill_super)(struct super_block *, void *, int))
1107 {
1108 	struct block_device *bdev;
1109 	struct super_block *s;
1110 	fmode_t mode = FMODE_READ | FMODE_EXCL;
1111 	int error = 0;
1112 
1113 	if (!(flags & SB_RDONLY))
1114 		mode |= FMODE_WRITE;
1115 
1116 	bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1117 	if (IS_ERR(bdev))
1118 		return ERR_CAST(bdev);
1119 
1120 	/*
1121 	 * once the super is inserted into the list by sget, s_umount
1122 	 * will protect the lockfs code from trying to start a snapshot
1123 	 * while we are mounting
1124 	 */
1125 	mutex_lock(&bdev->bd_fsfreeze_mutex);
1126 	if (bdev->bd_fsfreeze_count > 0) {
1127 		mutex_unlock(&bdev->bd_fsfreeze_mutex);
1128 		error = -EBUSY;
1129 		goto error_bdev;
1130 	}
1131 	s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC,
1132 		 bdev);
1133 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
1134 	if (IS_ERR(s))
1135 		goto error_s;
1136 
1137 	if (s->s_root) {
1138 		if ((flags ^ s->s_flags) & SB_RDONLY) {
1139 			deactivate_locked_super(s);
1140 			error = -EBUSY;
1141 			goto error_bdev;
1142 		}
1143 
1144 		/*
1145 		 * s_umount nests inside bd_mutex during
1146 		 * __invalidate_device().  blkdev_put() acquires
1147 		 * bd_mutex and can't be called under s_umount.  Drop
1148 		 * s_umount temporarily.  This is safe as we're
1149 		 * holding an active reference.
1150 		 */
1151 		up_write(&s->s_umount);
1152 		blkdev_put(bdev, mode);
1153 		down_write(&s->s_umount);
1154 	} else {
1155 		s->s_mode = mode;
1156 		snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1157 		sb_set_blocksize(s, block_size(bdev));
1158 		error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1159 		if (error) {
1160 			deactivate_locked_super(s);
1161 			goto error;
1162 		}
1163 
1164 		s->s_flags |= SB_ACTIVE;
1165 		bdev->bd_super = s;
1166 	}
1167 
1168 	return dget(s->s_root);
1169 
1170 error_s:
1171 	error = PTR_ERR(s);
1172 error_bdev:
1173 	blkdev_put(bdev, mode);
1174 error:
1175 	return ERR_PTR(error);
1176 }
1177 EXPORT_SYMBOL(mount_bdev);
1178 
kill_block_super(struct super_block * sb)1179 void kill_block_super(struct super_block *sb)
1180 {
1181 	struct block_device *bdev = sb->s_bdev;
1182 	fmode_t mode = sb->s_mode;
1183 
1184 	bdev->bd_super = NULL;
1185 	generic_shutdown_super(sb);
1186 	sync_blockdev(bdev);
1187 	WARN_ON_ONCE(!(mode & FMODE_EXCL));
1188 	blkdev_put(bdev, mode | FMODE_EXCL);
1189 }
1190 
1191 EXPORT_SYMBOL(kill_block_super);
1192 #endif
1193 
mount_nodev(struct file_system_type * fs_type,int flags,void * data,int (* fill_super)(struct super_block *,void *,int))1194 struct dentry *mount_nodev(struct file_system_type *fs_type,
1195 	int flags, void *data,
1196 	int (*fill_super)(struct super_block *, void *, int))
1197 {
1198 	int error;
1199 	struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1200 
1201 	if (IS_ERR(s))
1202 		return ERR_CAST(s);
1203 
1204 	error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1205 	if (error) {
1206 		deactivate_locked_super(s);
1207 		return ERR_PTR(error);
1208 	}
1209 	s->s_flags |= SB_ACTIVE;
1210 	return dget(s->s_root);
1211 }
1212 EXPORT_SYMBOL(mount_nodev);
1213 
compare_single(struct super_block * s,void * p)1214 static int compare_single(struct super_block *s, void *p)
1215 {
1216 	return 1;
1217 }
1218 
mount_single(struct file_system_type * fs_type,int flags,void * data,int (* fill_super)(struct super_block *,void *,int))1219 struct dentry *mount_single(struct file_system_type *fs_type,
1220 	int flags, void *data,
1221 	int (*fill_super)(struct super_block *, void *, int))
1222 {
1223 	struct super_block *s;
1224 	int error;
1225 
1226 	s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1227 	if (IS_ERR(s))
1228 		return ERR_CAST(s);
1229 	if (!s->s_root) {
1230 		error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1231 		if (error) {
1232 			deactivate_locked_super(s);
1233 			return ERR_PTR(error);
1234 		}
1235 		s->s_flags |= SB_ACTIVE;
1236 	} else {
1237 		do_remount_sb(s, flags, data, 0);
1238 	}
1239 	return dget(s->s_root);
1240 }
1241 EXPORT_SYMBOL(mount_single);
1242 
1243 struct dentry *
mount_fs(struct file_system_type * type,int flags,const char * name,void * data)1244 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1245 {
1246 	struct dentry *root;
1247 	struct super_block *sb;
1248 	char *secdata = NULL;
1249 	int error = -ENOMEM;
1250 
1251 	if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1252 		secdata = alloc_secdata();
1253 		if (!secdata)
1254 			goto out;
1255 
1256 		error = security_sb_copy_data(data, secdata);
1257 		if (error)
1258 			goto out_free_secdata;
1259 	}
1260 
1261 	root = type->mount(type, flags, name, data);
1262 	if (IS_ERR(root)) {
1263 		error = PTR_ERR(root);
1264 		goto out_free_secdata;
1265 	}
1266 	sb = root->d_sb;
1267 	BUG_ON(!sb);
1268 	WARN_ON(!sb->s_bdi);
1269 
1270 	/*
1271 	 * Write barrier is for super_cache_count(). We place it before setting
1272 	 * SB_BORN as the data dependency between the two functions is the
1273 	 * superblock structure contents that we just set up, not the SB_BORN
1274 	 * flag.
1275 	 */
1276 	smp_wmb();
1277 	sb->s_flags |= SB_BORN;
1278 
1279 	error = security_sb_kern_mount(sb, flags, secdata);
1280 	if (error)
1281 		goto out_sb;
1282 
1283 	/*
1284 	 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1285 	 * but s_maxbytes was an unsigned long long for many releases. Throw
1286 	 * this warning for a little while to try and catch filesystems that
1287 	 * violate this rule.
1288 	 */
1289 	WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1290 		"negative value (%lld)\n", type->name, sb->s_maxbytes);
1291 
1292 	up_write(&sb->s_umount);
1293 	free_secdata(secdata);
1294 	return root;
1295 out_sb:
1296 	dput(root);
1297 	deactivate_locked_super(sb);
1298 out_free_secdata:
1299 	free_secdata(secdata);
1300 out:
1301 	return ERR_PTR(error);
1302 }
1303 
1304 /*
1305  * Setup private BDI for given superblock. It gets automatically cleaned up
1306  * in generic_shutdown_super().
1307  */
super_setup_bdi_name(struct super_block * sb,char * fmt,...)1308 int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1309 {
1310 	struct backing_dev_info *bdi;
1311 	int err;
1312 	va_list args;
1313 
1314 	bdi = bdi_alloc(GFP_KERNEL);
1315 	if (!bdi)
1316 		return -ENOMEM;
1317 
1318 	bdi->name = sb->s_type->name;
1319 
1320 	va_start(args, fmt);
1321 	err = bdi_register_va(bdi, fmt, args);
1322 	va_end(args);
1323 	if (err) {
1324 		bdi_put(bdi);
1325 		return err;
1326 	}
1327 	WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1328 	sb->s_bdi = bdi;
1329 
1330 	return 0;
1331 }
1332 EXPORT_SYMBOL(super_setup_bdi_name);
1333 
1334 /*
1335  * Setup private BDI for given superblock. I gets automatically cleaned up
1336  * in generic_shutdown_super().
1337  */
super_setup_bdi(struct super_block * sb)1338 int super_setup_bdi(struct super_block *sb)
1339 {
1340 	static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1341 
1342 	return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1343 				    atomic_long_inc_return(&bdi_seq));
1344 }
1345 EXPORT_SYMBOL(super_setup_bdi);
1346 
1347 /*
1348  * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1349  * instead.
1350  */
__sb_end_write(struct super_block * sb,int level)1351 void __sb_end_write(struct super_block *sb, int level)
1352 {
1353 	percpu_up_read(sb->s_writers.rw_sem + level-1);
1354 }
1355 EXPORT_SYMBOL(__sb_end_write);
1356 
1357 /*
1358  * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1359  * instead.
1360  */
__sb_start_write(struct super_block * sb,int level,bool wait)1361 int __sb_start_write(struct super_block *sb, int level, bool wait)
1362 {
1363 	if (!wait)
1364 		return percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1365 
1366 	percpu_down_read(sb->s_writers.rw_sem + level-1);
1367 	return 1;
1368 }
1369 EXPORT_SYMBOL(__sb_start_write);
1370 
1371 /**
1372  * sb_wait_write - wait until all writers to given file system finish
1373  * @sb: the super for which we wait
1374  * @level: type of writers we wait for (normal vs page fault)
1375  *
1376  * This function waits until there are no writers of given type to given file
1377  * system.
1378  */
sb_wait_write(struct super_block * sb,int level)1379 static void sb_wait_write(struct super_block *sb, int level)
1380 {
1381 	percpu_down_write(sb->s_writers.rw_sem + level-1);
1382 }
1383 
1384 /*
1385  * We are going to return to userspace and forget about these locks, the
1386  * ownership goes to the caller of thaw_super() which does unlock().
1387  */
lockdep_sb_freeze_release(struct super_block * sb)1388 static void lockdep_sb_freeze_release(struct super_block *sb)
1389 {
1390 	int level;
1391 
1392 	for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1393 		percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1394 }
1395 
1396 /*
1397  * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1398  */
lockdep_sb_freeze_acquire(struct super_block * sb)1399 static void lockdep_sb_freeze_acquire(struct super_block *sb)
1400 {
1401 	int level;
1402 
1403 	for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1404 		percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1405 }
1406 
sb_freeze_unlock(struct super_block * sb,int level)1407 static void sb_freeze_unlock(struct super_block *sb, int level)
1408 {
1409 	for (level--; level >= 0; level--)
1410 		percpu_up_write(sb->s_writers.rw_sem + level);
1411 }
1412 
1413 /**
1414  * freeze_super - lock the filesystem and force it into a consistent state
1415  * @sb: the super to lock
1416  *
1417  * Syncs the super to make sure the filesystem is consistent and calls the fs's
1418  * freeze_fs.  Subsequent calls to this without first thawing the fs will return
1419  * -EBUSY.
1420  *
1421  * During this function, sb->s_writers.frozen goes through these values:
1422  *
1423  * SB_UNFROZEN: File system is normal, all writes progress as usual.
1424  *
1425  * SB_FREEZE_WRITE: The file system is in the process of being frozen.  New
1426  * writes should be blocked, though page faults are still allowed. We wait for
1427  * all writes to complete and then proceed to the next stage.
1428  *
1429  * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1430  * but internal fs threads can still modify the filesystem (although they
1431  * should not dirty new pages or inodes), writeback can run etc. After waiting
1432  * for all running page faults we sync the filesystem which will clean all
1433  * dirty pages and inodes (no new dirty pages or inodes can be created when
1434  * sync is running).
1435  *
1436  * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1437  * modification are blocked (e.g. XFS preallocation truncation on inode
1438  * reclaim). This is usually implemented by blocking new transactions for
1439  * filesystems that have them and need this additional guard. After all
1440  * internal writers are finished we call ->freeze_fs() to finish filesystem
1441  * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1442  * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1443  *
1444  * sb->s_writers.frozen is protected by sb->s_umount.
1445  */
freeze_super(struct super_block * sb)1446 int freeze_super(struct super_block *sb)
1447 {
1448 	int ret;
1449 
1450 	atomic_inc(&sb->s_active);
1451 	down_write(&sb->s_umount);
1452 	if (sb->s_writers.frozen != SB_UNFROZEN) {
1453 		deactivate_locked_super(sb);
1454 		return -EBUSY;
1455 	}
1456 
1457 	if (!(sb->s_flags & SB_BORN)) {
1458 		up_write(&sb->s_umount);
1459 		return 0;	/* sic - it's "nothing to do" */
1460 	}
1461 
1462 	if (sb_rdonly(sb)) {
1463 		/* Nothing to do really... */
1464 		sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1465 		up_write(&sb->s_umount);
1466 		return 0;
1467 	}
1468 
1469 	sb->s_writers.frozen = SB_FREEZE_WRITE;
1470 	/* Release s_umount to preserve sb_start_write -> s_umount ordering */
1471 	up_write(&sb->s_umount);
1472 	sb_wait_write(sb, SB_FREEZE_WRITE);
1473 	down_write(&sb->s_umount);
1474 
1475 	/* Now we go and block page faults... */
1476 	sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1477 	sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1478 
1479 	/* All writers are done so after syncing there won't be dirty data */
1480 	ret = sync_filesystem(sb);
1481 	if (ret) {
1482 		sb->s_writers.frozen = SB_UNFROZEN;
1483 		sb_freeze_unlock(sb, SB_FREEZE_PAGEFAULT);
1484 		wake_up(&sb->s_writers.wait_unfrozen);
1485 		deactivate_locked_super(sb);
1486 		return ret;
1487 	}
1488 
1489 	/* Now wait for internal filesystem counter */
1490 	sb->s_writers.frozen = SB_FREEZE_FS;
1491 	sb_wait_write(sb, SB_FREEZE_FS);
1492 
1493 	if (sb->s_op->freeze_fs) {
1494 		ret = sb->s_op->freeze_fs(sb);
1495 		if (ret) {
1496 			printk(KERN_ERR
1497 				"VFS:Filesystem freeze failed\n");
1498 			sb->s_writers.frozen = SB_UNFROZEN;
1499 			sb_freeze_unlock(sb, SB_FREEZE_FS);
1500 			wake_up(&sb->s_writers.wait_unfrozen);
1501 			deactivate_locked_super(sb);
1502 			return ret;
1503 		}
1504 	}
1505 	/*
1506 	 * For debugging purposes so that fs can warn if it sees write activity
1507 	 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1508 	 */
1509 	sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1510 	lockdep_sb_freeze_release(sb);
1511 	up_write(&sb->s_umount);
1512 	return 0;
1513 }
1514 EXPORT_SYMBOL(freeze_super);
1515 
1516 /**
1517  * thaw_super -- unlock filesystem
1518  * @sb: the super to thaw
1519  *
1520  * Unlocks the filesystem and marks it writeable again after freeze_super().
1521  */
thaw_super_locked(struct super_block * sb)1522 static int thaw_super_locked(struct super_block *sb)
1523 {
1524 	int error;
1525 
1526 	if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
1527 		up_write(&sb->s_umount);
1528 		return -EINVAL;
1529 	}
1530 
1531 	if (sb_rdonly(sb)) {
1532 		sb->s_writers.frozen = SB_UNFROZEN;
1533 		goto out;
1534 	}
1535 
1536 	lockdep_sb_freeze_acquire(sb);
1537 
1538 	if (sb->s_op->unfreeze_fs) {
1539 		error = sb->s_op->unfreeze_fs(sb);
1540 		if (error) {
1541 			printk(KERN_ERR
1542 				"VFS:Filesystem thaw failed\n");
1543 			lockdep_sb_freeze_release(sb);
1544 			up_write(&sb->s_umount);
1545 			return error;
1546 		}
1547 	}
1548 
1549 	sb->s_writers.frozen = SB_UNFROZEN;
1550 	sb_freeze_unlock(sb, SB_FREEZE_FS);
1551 out:
1552 	wake_up(&sb->s_writers.wait_unfrozen);
1553 	deactivate_locked_super(sb);
1554 	return 0;
1555 }
1556 
thaw_super(struct super_block * sb)1557 int thaw_super(struct super_block *sb)
1558 {
1559 	down_write(&sb->s_umount);
1560 	return thaw_super_locked(sb);
1561 }
1562 EXPORT_SYMBOL(thaw_super);
1563